California professors explain their shift to paper exams as AI clouds homework grades

A correct homework answer is no longer enough for some California professors to judge what a student knows. In newly published New York Post interviews, instructors described moving grades toward paper exams without devices, or planning that shift. Their responses separate two questions: whether AI can help students learn, and whether submitted work proves they have learned.

The answer no longer identifies the author

UC Berkeley mathematician Richard Borcherds told the Post he now bases grades mainly on in-person exams with no electronic devices.

Borcherds said average scores rise substantially when exams move online, which he attributes to students using AI. But he also acknowledged that he cannot tell which students used it.

In follow-up coverage, Tom’s Hardware quoted Borcherds arguing that grading homework for correct answers rewards AI use and penalizes honesty. He also called paying people to grade AI-generated answers a waste of resources. His objection concerns what a submitted answer can establish about the student, not simply whether that answer is right.

It’s become increasingly difficult to know what anyone knows and can do, and that signal is important.

Morgan Polikoff, USC education professor, speaking to the New York Post

Homework remains, with a different grading rule

At UCLA, statistics and data science lecturer Thomas Maierhofer described a clear division. Exams take place in person, on paper, without phones or laptops. Homework and labs completed outside class are graded for completion. The approach keeps those assignments in the course without treating their correctness as the same evidence as an exam.

Maierhofer sees AI as a potential teaching aid, not something to exclude from every learning activity. His argument is that device-free exams leave students accountable for understanding the material: using AI to avoid learning would hurt them when they must answer on their own.

USC education professor Morgan Polikoff described a planned change rather than one already implemented. Currently on leave from teaching, he said he had intended to move more graded undergraduate assignments to sit-down, paper-and-pencil tests. He expected to adopt a policy similar to Borcherds’s and argued for more assessments resistant to AI substitution.

Polikoff also recounted confronting a freshman two years ago over an AI-written paper, despite his ban on its use. The student admitted it, he said, and had to rewrite the paper from scratch.

AI support does not mean accepting AI-written work

USC allows instructors to set course-specific AI rules. Polikoff said his PhD program was debating a policy, but that it would still leave decisions to faculty.

Stanford Law professor Julian Nyarko offered a more favorable view of AI’s teaching potential, particularly compared with poor human instruction. Yet he warned against having it simply write a paper or solve a question for a student. He has always used in-class exams and said most colleagues who once assigned long take-home exams now feel unable to do so because of AI.